Walking by a billboard ad in the future might be a bit like stepping on a scale -- and your weight is valuable for advertisers.
Computer scientists created a system so digital billboards can remotely detect the weights of people passing by.
A camera with depth perception provides a weight estimate that could be used to better target advertisements.
With this sensing capability, physical advertisements could get far more personalized.
Digital ads in public spaces (subways, bus shelters, and airports) aren't yet on a first-name basis with passers-by, but walking by an ad and seeing your name pop up is getting closer to reality.
Now, computer scientists are working on ads that have the ability to read your weight. Researchers in Singapore have created a system that allows ads to accurately estimate the heft of people passing by.
"There exists a variety of practical situations where having an estimate of human weight is extremely useful," said Shuicheng Yan, an assistant professor of electrical and computer engineering at the National University of Singapore who led the weight-detection system development.
Cameras embedded in the ads would "read" weights so the ads could change based on the demographics in the area. Advertisers' goal is to accumulate enough weight data to customize the ads.
For example, if ad sense a number of athletic women passing by a camera-equipped ad, ads would be modified for that demographic.
"You will see a suitable advertisement -- clothing, make-up products, shops which are suitable for your height, weight, and age -- according to your demographic information," Yan said.
Or, if the camera senses an overweight population, the location might be ideal for advertising healthful foods or exercise-related products.
Facial recognition technology on billboard ads that collects age and gender data already exists, so Yan and his colleagues focused on weight as another potentially valuable piece of information.
The technology uses a depth camera similar to the kind found in an Xbox Kinect to sense and determine the "volume" of a passing human. Once the camera estimates the volume, the computer scientists use algorithms for human body detection and body feature extraction to figure out what the person weighs. That data is also run through a standard body mass index.
So far, the system has been tested in the lab on 200 participants in Singapore who ranged from about 90 to 220 pounds. The detected weight appeared on a computer screen set up in the lab so it could be checked against each participant's actual weight. Yan envisions the system being incorporated seamlessly into the back end of digital advertisements.
"All the technology companies are pushing the boundaries of your tolerance for privacy. So we'll see what happens," Kai Yu, the head of media analytics department at NEC Labs in Silicon Valley, said.
Besides bus terminals or areas where people have to wait and stay stationed for a time, the technology could be used in the gaming world. Conventionally, avatars are personalized with facial designs, but with this technology the player's weight and height could be applied to the avatar to make it more realistic.
Recently Yan's team demonstrated the system in Barcelona at the International Conference on Computing Vision. He and his colleagues submitted a paper describing the technology to the journal IEEE Transactions on Image Processing, where it is under review.
The computer scientists plan to add more data to the system so it can handle a wider variety of human poses and layers of clothing. Once the scientists finalize the technology, they want to combine the weight-detecting capabilities with other cues such as age, gender, ethnicity, height, and style of dress to create intelligent recommendations that advertisers can use, Yan said.
With weight-estimating capabilities, Yan says, you can infer more information about a person's profile, Yu said. "Then you can adapt the content to make the targeted advertisement more accurate."
By adding facial recognition software to the mix, advertisers could theoretically create hyper-personalized ads that can call viewers by name, but that Minority Report-like capability might creep people out.
David Fleet, a computer science professor at the University of Toronto who specializes in computer vision, emphasizes that this latest innovation is just one piece of information that can be gathered. Other characteristics are much more subtle.
"It's known, for example, that certain types of clinical depression can be detected in the way that people move," Fleet said. "There are going to be all kinds of sensitive medical, emotional information extracted from images of people."
Detecting personal characteristics using cameras will be significant, no matter who creates the applications, he added.
"The concerns we raise do not begin with this technology, it begins with the pervasive use of surveillance systems in society," Fleet said.